gpio-mvebu: patch for separate edge and level mask caches

[Goran Cengic]

Including the patch and lakml.

> -----Original Message-----
> From: Jason Cooper [mailto:jason at lakedaemon.net]
> Sent: den 7 oktober 2015 13:41
> To: Goran Cengic; Linus Walleij
> Cc: Thomas Petazzoni; Andrew Lunn; Gregory Clement; Sebastian
> Hesselbarth
> Subject: Re: gpio-mvebu: patch for separate edge and level mask caches
> 
> Hello Goran,
> 
> +Cc: LinusW (gpio maintainer)
> 
> On Wed, Oct 07, 2015 at 10:36:01AM +0000, Goran Cengic wrote:
> > Hi again,
> >
> > I didn't found who maintains the drivers/gpio/gpio-mvebu.c in the
> > MAINTAINERS and I've realized that Armada 38x SoC maintainers also
> > might be relevant for this issue so I've included them here.
> 
> You're absolutely correct.  Those of us involved with kernel development
> day-to-day 'just know' that the mvebu maintainers would handle such a
> driver.  We should update MAINTAINERS.
> 
> 
> > From: Goran Cengic
> > Sent: den 5 oktober 2015 20:15
> > To: 'Thomas Petazzoni'
> > Subject: gpio-mvebu: patch for separate edge and level mask caches
> >
> > Hi Thomas,
> >
> > We've encountered a small issue with not being able to mix edge and
> > level sensitive interrupts. The issue is that gpio-mvebu would trigger
> > an edge sensitive interrupt even though mvebu_gpio_level_irq_mask()
> > has been called for the same level interrupt. The issue is caused by
> > the mask_cache in irq_chip_generic being shared between the
> > chip_types.
> >
> > Here is a summary of what happens. Enable_irq(some level sensitive
> > interrupt), as configured in dtb, enable_irq(some other edge sensitive
> > interrupt). At this point the mask_cache is not masking bot
> > interrupts. Since the same mask_cache is written to the edge mask
> > register, at the second call to enable_irq(), the level sensitive
> > interrupt is unmasked in the edge mask register also. Now as soon as
> > the transition to, say, low level occurs the mvebu gpio controller
> > registers one in the edge cause register while the data in is used as
> > level cause register. The irq handling starts and masks the interrupt
> > using mvebu_gpio_level_irq_mask(), since that interrupt is configured
> > as level-sensitive. However, the edge cause register is never
> > mvebu_gpio_irq_ack()'ed and continues to trigger interrupt handling
> > and locks the processor since the edge mask register is not masking
> > that bit.
> 
> Thanks for the bug report, however ...
> 
> > I've attached a small patch so that you can decide if it's ok for inclusion.
> 
> hmmm, I don't see it.  Could you please resend with it attached?
> 
> Please also Cc lakml at linux-arm-kernel at lists.infradead.org.  If you can, then
> please try to adhere to the list conventions: no html, no legal footers, etc.
> 
> thx,
> 
> Jason.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-gpio-mvebu-separate-edge-and-level-mask-caches.patch
Type: application/octet-stream
Size: 2173 bytes
Desc: 0001-gpio-mvebu-separate-edge-and-level-mask-caches.patch
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20151007/5d4715d9/attachment-0001.obj>